Freepik releases an ‘open’ AI image generator trained on licensed data

Freepik, online graphic design platform presented a new “Open” AI image model on Tuesday that the company was taught only by commercial licensed, “safe work” images.

Model, called F liteContains about 10 billion parameters – Internal components that make up the Settings model. According to Freepik, according to FREE, F Lite AI is developed in common with Fal.Alat, and using 64 NVidia H100 GPUs, according to Freepik.

F Lite joins a small and growing collection of generative AI models taught licensed data.

We are secretly working on this for months! Finally feels good to share!

Links:

• Permanent version: more predictable and tidy-loyal, but less artistic: https://t.co/mywsker9ir

• Docker version: more chaotic and wrong inclined, but better tissues and … pic.twitter.com/gx5mipye8o

– Report Lopez ⛩️ (@javilopen) 29 April 2025

Generative AI is in the center of copyright claims against EU companies Open and Midjourney. Often developed using mass content – copyrighted content – public sources around the Internet. Most companies developing these models are arguing Use the exhibition The owners protect their experience using copyrighted data for training without compensating. Many creative and IP law owners do not agree.

Freepik, both internal data sets made up of about 80 million images, F Lite, which has a standard and texture made in the image, made two flavorings of the standard and tissue. The standard is more predictable and “aggressive,” the texture is more “wrong tendency”, but the company presents better tissues and creative compositions.

Here is an image of a standard model created with “a person who stands in front of a sunset, a majestic surrounding person.”

FreePik, F Lite, Midjourney’s V7 does not claim to create superior images from leading image generators such as black forest labs’ flux family or others. The goal was to prepare a model open to adapt and develop developers, according to the company.

This is not an easy feat to work, FITI. The model requires a GPU with at least 24GB VRAM.

Includes other companies developing media-creating models in licensed data Adobe, Bria, Getty pictures, Moonvalleyand silence. Depending on how the AI ​​copies of copyright claims can grow in the market exponentially.